MoGAIT — Motion & Gait AI Intelligence Tool

AI-powered decision support system for rehabilitation robotics and gait analysis

Developed for the AI Hackathon for People with Disabilities
King Salman Center for Disability Research · Riyadh · June 14, 2026

Prof. Dr. Utku Köse · IEEE Senior Member · ACM Professional Member
Süleyman Demirel University · University of North Dakota · VelTech University · Universidad Panamericana

Overview

MoGAIT is a single-file, browser-based AI platform that integrates neural network-driven patient synthesis, explainable AI (XAI), rehabilitation robotics decision support, and multi-scenario mobility simulation — all without a server, database, or installation.

It was designed to answer a practical question at the intersection of AI and disability research: "Given this patient's profile, in this specific real-world context, which rehabilitation robot should be used, how should it be controlled, and what outcome can be expected?"

The system addresses clinical, community, and mass-gathering mobility scenarios — including Hajj pilgrimage, post-stroke gait rehabilitation, paediatric mobility, and 13 other real-world contexts.

Live Demo

▶ Open MoGAIT in your browser

No installation. No sign-up. Works on Chrome, Edge, Firefox (desktop).

Quick start:

Enter a patient count (e.g. 30) → click Generate
Select any scenario from the strip (try Hajj / Pilgrimage or Gait Recovery)
Click any patient card on the left to open their full analysis
Switch to Robotics & Simulation tab for device matching, simulation, and planning

Key Features

Module	What it does
Synthetic Cohort Generation	VAE+MLP generates realistic patient cohorts (10–200 patients) across 8 diagnosis groups
Multi-Domain Risk Profiling	6-dimension risk profile: Fall Risk, Fatigue, Pain Impact, Mobility Score, Participation Index, Rehab Potential
Scenario-Adaptive XAI	Feature importance re-weighted per scenario — same patient, different scenario = different XAI priorities
Counterfactual XAI	Interactive sliders simulate clinical interventions: "If we improve gait speed to 0.9 m/s, risk drops by X"
Device Match	6 robot types matched to patient by diagnosis, scenario, and clinical indicators
Control Dynamics	6 control algorithms visualised as patient-specific torque profiles over the gait cycle
Gait-Robot Interaction	Pre/post therapy hip flexion comparison with diagnosis-stratified RCT effect sizes
4-Week Session Plan	Personalised rehabilitation plan with weekly targets, session intensity, and dual-axis progression chart
Cohort Simulation	40-agent without-robot vs. with-robot comparative animation across 13 environments
Patient Simulation	Selected patient starred (★) in both panels; live fall count, speed gain, and fatigue reduction
AI Clinical Advisor	Claude, Gemini, or GPT-4o generates scenario-aware clinical reasoning (BYOK)
19 Scenarios	13 primary + 6 faith & cultural gathering scenarios, each with independent XAI weights

AI Architecture

Patient Generator (GEN_NET)

Latent vector z ∈ ℝ¹²  →  [32] → [48] → [32] → [20]  →  Patient parameters

A VAE-style decoder generates 20 clinical parameters from a 12-dimensional random latent input. Each of the 8 diagnosis groups (stroke, SCI, CP, MS, amputation, elderly, TBI, Parkinson's) has group-specific priors for gait speed, asymmetry, variability, comorbidity burden, and energy cost. The network uses ReLU activations with sigmoid/tanh output clamping per parameter type.

Generated parameters include:

Gait speed (m/s), step asymmetry (%), gait variability (CV%), cadence (spm)
TUG test time (sec), stride length (m), double support time (%)
Fall history (falls/year), pain score (VAS 0–10), energy cost (J/kg/m)
Medications count, comorbidity list, age, BMI, SpO₂, heart rate
Post-event week, WMCA classification

Risk Classifier (RISK_NET)

10 clinical inputs  →  [24] → [24] → [16] → [4]  →  Risk tier (Low / Moderate / High / Critical)

A 4-layer MLP classifies each patient into one of four risk tiers using 10 normalised inputs:

Input	Normalisation	Clinical weight
Gait Speed	÷ 1.4	Fall risk inverse
Step Asymmetry	÷ 70	Compensation marker
TUG Test	÷ 62	Mobility threshold
Fall History	÷ 10	Strongest predictor
Pain Score	÷ 10	Compensation driver
Medications	÷ 13	Polypharmacy risk
Comorbidities	÷ 10	Systemic burden
Gait Variability	÷ 17	Pre-fall signal
Diagnosis Severity	categorical	Neurological grade
Age (normalised)	÷ 90	Physiological reserve

A power-transform calibration (raw^0.72 × 88 + 5) prevents ceiling saturation and ensures a realistic clinical distribution (~30% low / ~40% moderate / ~30% high risk).

Scenario-Adaptive XAI Engine

Each feature's contribution to risk is measured via finite-difference gradient sensitivity:

∂risk/∂xᵢ ≈ [f(x + εeᵢ) − f(x − εeᵢ)] / 2ε,   ε = 0.001

These raw gradients are then re-weighted by the active scenario's SC_MODS vector — a 10-dimensional weight array that encodes which clinical features matter most in each scenario:

Scenario	Top amplified features	Clinical rationale
Fall Prevention	▲ Fall History (×1.6), ▲ TUG (×1.4)	Direct fall predictors
Exoskeleton Candidacy	▲ Diagnosis (×1.7), ▲ Age (×1.5)	Device eligibility criteria
Sports Rehabilitation	▲ Asymmetry (×1.7)	Limb Symmetry Index target
Hajj / Pilgrimage	▲ Age (×1.6), ▲ Variability (×1.5), ▲ Comorbidity (×1.5)	Heat + crowd collapse risk
Energy & Fatigue	▲ Variability (×1.4), ▲ Age (×1.5)	Fatigue compounds over time
Pain & Compensation	▲ Pain (×1.8), ▲ Asymmetry (×1.5)	Compensatory gait patterns

When a scenario is changed, the XAI panel updates instantly — the same patient shows different feature importances, with ▲ amber badges for amplified features and ▼ grey badges for depressed ones.

Counterfactual XAI

"If we improve this patient's gait speed from 0.64 m/s to 0.90 m/s via exoskeleton — how much does risk drop?"

Six interactive sliders allow simulation of clinical interventions. Each slider change re-runs RISK_NET instantly and displays the Δ risk score. Each slider is labelled with the intervention that would achieve it (e.g., "Exoskeleton, treadmill training, walker" for gait speed).

This implements counterfactual reasoning — XAI showing not just which features matter, but how much changing them would help.

Multi-LLM Clinical Advisor

Three large language models are available via user-supplied API keys (BYOK — keys used only for the direct API call, never stored):

Claude (Anthropic) — primary clinical reasoning
Gemini (Google)
GPT-4o (OpenAI)

The advisor receives the full patient profile, active scenario, XAI attribution results, and risk scores. It generates scenario-contextualised clinical interpretation including intervention priorities, safety flags, and monitoring plan.

Scenario System

MoGAIT supports 19 scenarios in two groups:

Primary Scenarios (always visible)

Scenario	XAI Focus	Environment
🛡️ Fall Prevention	Fall history, TUG, polypharmacy	Home obstacle course
🔄 Gait Recovery	Speed, asymmetry	Clinic parallel bars
🦾 Exoskeleton Candidacy	Diagnosis severity, age	Clinic corridor
⚡ Energy & Fatigue	Variability, age	Community walking
⚖️ Pain & Compensation	Pain, asymmetry	Clinic
🏘️ Community Participation	Speed, participation	Urban street
🧠 Neurological Rehab	Asymmetry, fall history	Clinic
🏥 Post-Surgical Mobility	Speed, polypharmacy	Hospital ward
🏋️ Sports Rehabilitation	Asymmetry (LSI)	Rehab gymnasium
🕌 Hajj / Pilgrimage	Age, variability, comorbidity	Tawaf (Kaaba orbit)
🚶 Daily Community Walking	Speed, participation	Community
🏭 Workplace Ergonomics	Pain, asymmetry	Factory/office
👶 Paediatric Mobility	Diagnosis, participation	Clinic

Additional: Faith & Cultural Gathering Scenarios

Accessible via the 🌍 Additional dropdown. All validated for robotic rehabilitation compatibility (flat terrain, controlled environments):

Scenario	Notes
🕍 Wailing Wall Plaza	Cobblestone, elderly pilgrims, hip-assist optimal
✝️ Vatican Pilgrimage	St. Peter Square, prolonged standing
🕋 Umrah (Off-peak)	Marble surface, off-peak Tawaf
🏔️ Lourdes Pilgrimage	Most accessible pilgrimage site globally
🛤️ Camino (Urban Stages)	Flat paved sections, FDA-accessible routes
🌿 Medjugorje Pilgrimage	Gentle terrain, disability pilgrim infrastructure

Note: Kumbh Mela was explicitly excluded — stampede risk, mud terrain, and no assistive robot infrastructure make it incompatible with exoskeleton use.

Patient Analysis Module

Cohort Overview

Generate 10–200 synthetic patients. The cohort overview shows:

Population summary: avg risk, high/moderate/low distribution, avg speed, TUG, falls/yr, robot benefit
Diagnosis distribution bar chart with per-diagnosis risk scores
Highest-risk patients for immediate triage
Search and filter by name or diagnosis group

Individual Patient View

Selecting a patient reveals:

Gait Parameters — Chart.js multi-axis chart showing Hip, Knee, and Ankle flexion angles across the gait cycle, with normative Winter (2009) reference curves and patient Left vs. Right (paretic) differentiation. Step asymmetry and variability are embedded directly in the curve shape.

Multi-Domain Risk Profile — 6-bar MLP-computed profile:

Fall Risk — RISK_NET output
Fatigue Risk — energy cost × age weighted
Pain Impact — VAS pain × compensation index
Mobility Score — ICF d4 proxy (speed × cadence)
Participation Index — ICF d9 proxy (ambulation safety)
Rehab Potential — neuroplasticity window × age

Clinical Flags & Comorbidities

Automated flagging of High TUG Score, Polypharmacy, Fall History, and comorbidities. Physiological markers include heart rate, SpO₂, energy cost (J/kg/m), and muscle activation (%MVC) for five muscle groups.

Evidence-Based Interventions

Scenario-adaptive intervention priorities with:

Priority level (Now / Soon)
Intervention category and specific recommendation
Full clinical rationale linked to patient parameters
Level of evidence (Cochrane Level I / RCT Level II)
In-line APA references

XAI Feature Attribution + 12-Week Trajectory

Horizontal bar chart of 10 feature importances (absolute gradient magnitude)
Scenario weight multipliers shown per bar (×1.7, ×0.6 etc.)
12-week trajectory: gait speed (left axis), fall risk + participation index (right axis)
Diagnosis-stratified recovery rates with plateau annotation
Age and comorbidity modifiers applied to recovery slope

Counterfactual XAI

AI Clinical Advisor

Robotics & Simulation Module

Device Match

Population-level analysis shows all 6 robot types with:

Eligible patient count + % of cohort
Projected speed gain, fall risk reduction, energy saving
Scenario compatibility badges: ✓ For this scenario (teal) / ✗ Not recommended (red)
Clinical indication, best scenarios, full verified APA reference

Individual patient view (Device Match tab) computes compatibility scores (0–100) based on diagnosis, gait speed, age, and clinical setting — sorted by suitability.

Supported robot types:

Robot	Example	Best for
🏥 Tethered Exoskeleton	Lokomat Pro	Stroke, SCI, TBI (inpatient)
🦾 Wearable Exoskeleton	EksoGT / ReWalk	SCI, stroke (community)
💠 Hip-Assist Robot	GEMS / EX1	Elderly, prolonged walking
🦿 Smart Orthosis	REEV DREEVEN	Foot drop, MS, CP
🔩 End-Effector Robot	Lokohelp / G-EO	Gait initiation, spasticity
⚙️ Functional Trainer	AlterG / PDSYS	Sports rehab, body-weight support

Control Dynamics

Six control algorithms are selectable via card UI, each producing a characteristically different torque profile over the gait cycle:

Algorithm	Torque characteristic	Complexity
Position Control	Rigid step shape at stance/swing transition	Low
Impedance Control	Smooth sinusoidal + damping term	Medium
Admittance Control	Phase-shifted, lags patient force	Medium
Adaptive Control	Decaying envelope as patient effort increases	High
AI Intent Control	Leads the movement by ~60 ms (predictive)	High
EMG-BMI Control	Irregular bursts, cortical-pathway driven	V.High

Interactive sliders control stiffness (Kd), damping (Bd), max torque, and patient effort — the Chart.js chart re-renders in real time. Patient-specific modulation: the torque curve reflects this patient's actual gait speed, asymmetry, and variability.

Gait-Robot Interaction

Pre-therapy and post-therapy hip flexion angle compared against the Winter (2009) normative reference, rendered as a Chart.js multi-line chart. Post-therapy curves are computed from diagnosis-stratified RCT effect sizes (Mehrholz 2017, Lee et al. 2023).

Six outcome cards show projected 4-week gains: speed, fall risk, energy cost, TUG, step asymmetry, and energy saving — all with delta values vs baseline.

4-Week Session Plan

A personalised rehabilitation plan with:

Week-by-week phase (Familiarisation → Guided Practice → Active Participation → Challenge & Transfer)
Session frequency, duration, control algorithm, speed target, robot assist %, patient effort %
Starting assist level personalised to risk tier (high-risk/SCI → 90%; standard → 75%)
Chart.js mixed chart: session duration as bars, gait speed and assist/effort as dual-axis lines

Cohort & Patient Simulation

Cohort Simulation — 40-agent side-by-side comparison:

LEFT: WITHOUT robot — agents coloured by fall risk severity (🔴 high ≥65, 🟡 moderate, 🟢 low)
RIGHT: WITH robot — agents coloured by diagnosis group; robot type icon above each agent
ROBOT ROSTER — live panel showing which patient uses which device
Live HUD: falls, avg speed, avg fatigue per panel
Live stats: falls prevented, speed gain, fatigue reduction, robot benefit factor

Patient Simulation — selected patient tracked individually:

Patient highlighted with ★ star in both canvases
Live 4-metric panel: Falls (no robot) / Falls (with robot) / Speed gain / Fatigue reduction
Robot type matched to diagnosis (SCI → EksoGT 🦾, Elderly → GEMS 💠, Stroke → Lokomat 🏥...)

13 simulation environments including: clinical parallel bars, rehabilitation gymnasium, Tawaf (counterclockwise Kaaba orbit), cobblestone plaza, community street, hospital ward, workplace floor.

Clinical Evidence Base

MoGAIT's parameters, effect sizes, and recommendations are grounded in peer-reviewed literature:

Evidence item	Source
Electromechanical gait training: OR 1.94 for independent walking	Mehrholz et al. (2017), Cochrane
Hip-assist robot reduces energy cost 15–25% in elderly	Lee et al. (2023), Scientific Reports
AI-assisted rehabilitation superiority in MSK disorders	Luo et al. (2025), Front. Bioengineering
Gait speed <0.8 m/s: limited community ambulation threshold	Fritz & Lusardi (2009), JGPT
Gait variability: fall prediction sensitivity 80%+	Hausdorff et al. (2001), Arch Phys Med
Step asymmetry >25%: hemiparetic compensation	Patterson et al. (2010), Arch Phys Med
Intent recognition in rehabilitation robots (systematic review)	Luo et al. (2024), Disabil Rehabil Assist Technol
Control algorithms review (2020–2024)	Rad & Brișan (2025), Applied Sciences
AI-driven rehabilitation robotics	Abbas et al. (2025), Cureus
Hajj 2024: 1,301 heat deaths, pilgrim age 54.98 ± 13.96 yrs	Alzaben et al. (2024), PLOS ONE
Hajj AI & crowd management	Alam et al. (2024), IEEE Access
Telestroke at Hajj 2023–2024	Alfurayh et al. (2025), Front. Neurology

Technical Architecture

Aspect	Detail
Deployment	Single HTML file (`index.html`) — no server, no build step, no dependencies to install
File size	~204 KB
Neural network	Pure JavaScript — GEN_NET (VAE decoder 12→32→48→32→20) + RISK_NET (MLP 10→24→24→16→4)
Chart engine	Chart.js 4.4.1 (loaded from cdnjs CDN)
Simulation	Canvas API — 40-agent animation, 13 environment renderers
XAI	Finite-difference gradient sensitivity, scenario-reweighted (SC_MODS 10-dim vector per scenario)
LLM integration	Direct browser → Anthropic / Google / OpenAI API calls (BYOK, keys never stored)
Data	All patient data is synthetically generated at runtime — no real patient data, no storage
Browser support	Chrome 90+, Edge 90+, Firefox 88+ (desktop)

Repository Structure

mogait/
├── index.html          # The entire application (single file)
├── README.md           # This file
├── LICENSE             # MIT
└── images/             # Screenshots for README
    ├── 1.png           # Cohort Overview
    ├── 2.png           # Patient — Gait Parameters
    ├── 3.png           # Evidence-Based Interventions
    ├── 4.png           # Clinical Flags & Physiological Markers
    ├── 5.png           # XAI Attribution & 12-Week Trajectory
    ├── 6.png           # Counterfactual XAI
    ├── 7.png           # AI Clinical Advisor
    ├── 8.png           # Robotics — Population Device Match
    ├── 9.png           # Robotics — Individual Device Cards
    ├── 10.png          # Cohort Simulation
    ├── 11.png          # Individual Device Compatibility Scores
    ├── 12.png          # Control Dynamics Chart
    ├── 13.png          # Gait-Robot Interaction
    ├── 14.png          # 4-Week Session Plan
    └── 15.png          # Patient Simulation

Running Locally

Since MoGAIT is a single HTML file, running it locally is trivial:

# Option 1 — just open the file
open index.html          # macOS
start index.html         # Windows
xdg-open index.html      # Linux

# Option 2 — serve locally (avoids any CORS warnings)
python3 -m http.server 8080
# then open http://localhost:8080

GitHub Pages (recommended for sharing)

Fork or clone this repository
Go to Settings → Pages
Source: Deploy from a branch → main → / (root)
Save — your live URL will be https://yourusername.github.io/mogait/

The file must be named index.html for GitHub Pages to serve it automatically.

API Keys (Optional)

The AI Clinical Advisor requires API keys to function. These are entirely optional — all other features work without them.

Click 🔑 API Keys in the scenario strip
Enter one or more keys:
- Claude: get from console.anthropic.com
- Gemini: get from aistudio.google.com
- GPT-4o: get from platform.openai.com
Keys are used only for the direct API call from your browser — never stored, never sent to any server other than the respective provider

Clinical Disclaimer

MoGAIT generates synthetic data for research and demonstration purposes only.

All patient profiles are AI-generated. Clinical outcomes are illustrative estimates derived from published literature effect sizes, not predictions for any individual patient. This tool has not been validated for clinical decision-making and should not replace qualified clinical judgment.

Always consult licensed rehabilitation clinicians, physiotherapists, and physicians before making any clinical decisions.

References

Abbas, G. H., Speksnijder, C., Ramnarain, D., Parmar, C., Parmar, A., Ahmad, S., & Pouwels, S. (2025). AI-driven rehabilitation robotics: Advancements in and impacts on patient recovery. Cureus, 17(10), e94273. https://doi.org/10.7759/cureus.94273
Alam, M. S., Akbar, M. S., Iqbal, M. M., Jian, P., & Xiong, G. (2024). Enhancing Hajj and Umrah rituals and crowd management through AI technologies: A comprehensive survey. IEEE Access, 12, 161084–161120. https://doi.org/10.1109/ACCESS.2024.3487923
Alfurayh, N., et al. (2025). Telestroke management during the Hajj seasons 2023–2024: Insights from SEHA Virtual Hospital. Frontiers in Neurology. https://doi.org/10.3389/fneur.2025.1573275
Alzaben, A., Almutairi, N., & Alharthi, M. (2024). Health risk behaviors and associated factors among Hajj 2024 pilgrims: A multinational cross-sectional study. PLOS ONE, 19(11), e0314729. https://doi.org/10.1371/journal.pone.0314729
Fritz, S., & Lusardi, M. (2009). White paper: "Walking speed: The sixth vital sign." Journal of Geriatric Physical Therapy, 32(2), 2–5. https://doi.org/10.1519/00139143-200932020-00002
Hausdorff, J. M., Rios, D. A., & Edelberg, H. K. (2001). Gait variability and fall risk in community-living older adults: A 1-year prospective study. Archives of Physical Medicine and Rehabilitation, 82(8), 1050–1056. https://doi.org/10.1053/apmr.2001.24893
Lee, S.-H., Kim, J., Lim, B., Lee, H.-J., & Kim, Y.-H. (2023). Exercise with a wearable hip-assist robot improved physical function and walking efficiency in older adults. Scientific Reports, 13(1), 7269. https://doi.org/10.1038/s41598-023-32335-8
Luo, S., Meng, Q., Li, S., & Yu, H. (2024). Research of intent recognition in rehabilitation robots: A systematic review. Disability and Rehabilitation: Assistive Technology, 19(4), 1307–1318. https://doi.org/10.1080/17483107.2023.2170477
Luo, Z., Zhang, Y., Liu, T., & Wang, W. (2025). Effectiveness of artificial intelligence-assisted rehabilitation for musculoskeletal disorders: A network meta-analysis. Frontiers in Bioengineering and Biotechnology, 13, 1660524. https://doi.org/10.3389/fbioe.2025.1660524
Mehrholz, J., Thomas, S., Werner, C., Kugler, J., Pohl, M., & Elsner, B. (2017). Electromechanical-assisted training for walking after stroke. Cochrane Database of Systematic Reviews, 2017(5), CD006185. https://doi.org/10.1002/14651858.CD006185.pub4
Patterson, K. K., Parafianowicz, I., Danells, C. J., Closson, V., Verrier, M. C., Staines, W. R., Black, S. E., & McIlroy, W. E. (2010). Gait asymmetry in community-ambulating stroke survivors. Archives of Physical Medicine and Rehabilitation, 91(2), 205–212. https://doi.org/10.1016/j.apmr.2009.10.023
Rad, A.-B., & Brișan, C. (2025). Control algorithms in robot-assisted rehabilitation: A systematic review. Applied Sciences, 15(16), 9184. https://doi.org/10.3390/app15169184
Rahmani, A. (2024). Artificial intelligence and its revolutionary role in physical and mental rehabilitation. BioMed Research International, 2024, 9554590. https://doi.org/10.1155/2024/9554590

Prof. Dr. Utku Köse
utkukose@sdu.edu.tr · ukose@up.edu.mx
utkukose.com · ORCID: 0000-0002-9652-6415

AI Hackathon for People with Disabilities · King Salman Center for Disability Research · June 14, 2026

Navigation Menu

Repository files navigation

MoGAIT — Motion & Gait AI Intelligence Tool

Table of Contents

Overview

Live Demo

Key Features

AI Architecture

Patient Generator (GEN_NET)

Risk Classifier (RISK_NET)

Scenario-Adaptive XAI Engine

Counterfactual XAI

Multi-LLM Clinical Advisor

Scenario System

Primary Scenarios (always visible)

Additional: Faith & Cultural Gathering Scenarios

Patient Analysis Module

Cohort Overview

Individual Patient View

Robotics & Simulation Module

Device Match

Control Dynamics

Gait-Robot Interaction

4-Week Session Plan

Cohort & Patient Simulation

Clinical Evidence Base

Technical Architecture

Repository Structure

Running Locally

GitHub Pages (recommended for sharing)

API Keys (Optional)

Clinical Disclaimer

References

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
Latest commit History 3 Commits
images		images
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
index.html		index.html

Folders and files

Latest commit

History

Repository files navigation

MoGAIT — Motion & Gait AI Intelligence Tool

Table of Contents

Overview

Live Demo

Key Features

AI Architecture

Patient Generator (GEN_NET)

Risk Classifier (RISK_NET)

Scenario-Adaptive XAI Engine

Counterfactual XAI

Multi-LLM Clinical Advisor

Scenario System

Primary Scenarios (always visible)

Additional: Faith & Cultural Gathering Scenarios

Patient Analysis Module

Cohort Overview

Individual Patient View

Robotics & Simulation Module

Device Match

Control Dynamics

Gait-Robot Interaction

4-Week Session Plan

Cohort & Patient Simulation

Clinical Evidence Base

Technical Architecture

Repository Structure

Running Locally

GitHub Pages (recommended for sharing)

API Keys (Optional)

Clinical Disclaimer

References

About

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages